AI product UX is the design of what users can understand, inspect, approve, correct, and recover when a model participates in a task. These four patterns come from published Rahmanef product work, with links to the actual systems behind each example.
Rahman Fakhru · Field notes from product work
The examples describe implementation choices in specific products. They are prompts for evaluating a new brief, not universal safety guarantees or measured client outcomes.
When AI produces a draft, keep the human decision in the same workflow. Show who can review, what state the work is in, and what action moves it toward delivery.
Content Alfa joins brand context, production, share-link client review, and scheduling. AI assists the draft; approval and publication remain explicit decisions. Inspect Content Alfa’s approval workflow
A product should still explain its next step when a provider or integration is unavailable. Manual input and visible provider choice let users keep working and assess the cost of AI assistance.
Content Alfa documents BYOK alongside manual inputs, CSV, screenshots, and embeds. These are product decisions, not a claim that every external API works for every account. See Content Alfa’s BYOK and fallback decisions
An agent-facing surface needs a stable data representation and explicit access rules. A chat interface alone cannot define what an agent may read or change.
Open Silong exposes an MCP HTTP surface using structured JSON alongside its editor, databases, and collaboration model. The case study distinguishes integration from blanket access. Inspect Open Silong’s integration design
Test the user's path through interface, data, permissions, failure, and recovery. A polished screen does not prove that an operational task can be finished.
Open Silong grew as connected editor, database, sharing, and MCP slices. Content Alfa connects planning, drafting, review, and delivery instead of treating a generator as the entire product. See the product design process
For a new product, identify the user, the information the system may use, the actions that require consent, the fallback when AI is unavailable, and the evidence for acceptance. The interface and technical architecture should make those decisions visible.